"high ISO noise" vs noise revealed by the Exposure or Shadows slider??

[twalp]

This may be a dumb question but I'm baffled.

I'm working in CS6 on some RAW images that were shot on a D700 at ISO200 and are a bit underexposed. When I open up shadows (+21) and increase exposure slightly (+1.65) in CS6 ACR noise is introduced.  I tried adjusting the Luminance and Color sliders in ACR's Noise Reduction panel but I don't like the loss of detail. Plus, I read that using a dedicated noise reduction plugin offers superior results.

So now I've tried both Neat Image Pro and Topaz DeNoise. Using these plugins to reduce (not eliminate) the noise in my images requires adjustments that to my eye offer little benefit over ACR's Noise Reduction panel.  Detail is still lost.  The documentation for both plugins focuses on dealing with noise from shooting at high ISOs. Since neither plugin does much better on reducing the noise induced (or revealed?) by my shadows and exposure changes in ACR -- without killing detail --  I get the feeling that "my noise" isn't the same thing as the noise these plugins are designed for.  (I had assumed that the noise I'm fighting is still "device noise" from the D700, but just not evident in the original exposure.)

Here is a 300% screen shot of a portion of the image to show the noise I'm dealing with.  



Using any of the aforementioned tools I can get rid of the noise in the green chair and olive-colored metal wall, but then the details in the man's "hairy leg" are lost, which is illustrative of all the detail lost in faces, etc., elsewhere in the image. If I later add a little sharpening this just brings back noise or introduces new artifacts.

Is there no magic bullet?  

I'd appreciate some advice.

Thank you.

[Slobodan Blagojevic]

The attachment is missing, but anyway, I think I know what you are talking about. Post-processing is not introducing noise, just revealing it. It is most visible in underexposed shadow areas, so a solution would be to use a mask in PS to apply noise reduction only to those shadows.

[twalp]

Thank you, Slobodan.  I can't get the attachment function to work so I inserted the image in my post.

I appreciate your suggestion. My "Plan B" was to either paint in sharpening or noise reduction but I first wanted to perhaps understand why these noise reduction plugins didn't produce the hoped for results.  In Neat Image I loaded the profile for the D700 at 200 ISO but that had NO effect on the image, which is when I started to wonder if the noise I'm fighting has anything to do with the camera's inherent noise. I've written them and maybe they'll respond. But in the meantime I'm seeking both enlightenment about noise and a solution to my immediate problem. If local (vs global) correction is the only solution I'll indeed go that direction.

Thank you again for taking your time to respond. Much appreciated.

[Slobodan Blagojevic]

... why these noise reduction plugins didn't produce the hoped for results.  In Neat Image I loaded the profile for the D700 at 200 ISO but that had NO effect on the image...

Profiles are built around normally/properly exposed images. In the case of D700 profile at 200 ISO, the profile would count on rather minimal, if any noise - that is why you do not see any effect on the image). The noise you are fighting is underexposed noise, which may (or not) have a different pattern than a "normal" noise (this is above my "pay grade" - we have experts here on the forum that might clarify this).

For practical purposes, do not rely on built-in profiles, but go manually. Certain noise reduction programs (and I used Noise Ninja and Noiseware in the past) would allow you to modify the algorithm based on color, area frequency, tonality, etc. So, for instance, in your example, you might want to use a stronger noise reduction for green color than for orange one. Or you might chose shadows vs. midtones and highlights. Or high-frequency (small detail) to be affected less than low frequency. Or any combination of those, but a manual selection is the best way to go, depending on the image.

Check this example from our member DigitalDog (the whole debate is in this thread):

http://digitaldog.net/files/100vs800iso.jpg

You will see that, under certain circumstances, an 800 ISO image might have less noise than a 100 ISO image, making a ready-made noise reduction profile practically useless.

[Keith Reeder]

Is there no magic bullet?   

There is.

Stop looking at your files at 300%.

Seriously - nothing good ever comes of pixel-peeping.

[RFPhotography]

I'd suggest that 1 2/3 stops underexposed is more than 'slight' underexposure.  

You're dealing with a pretty seriously underexposed image and noise will become very evident if you try to push things up.

All noise reduction algorithms are going to cause a loss of detail of some magnitude.  It's the nature of the beast and in the end it's a tradeoff between detail and noise reduction.  Are you prepared to sacrifice detail for less noise or are you prepared to live with more noise for better detail.

You can also try applying noise reduction on a single channel.  Since what you're dealing with is in the green/olive family, you may be able to effectively isolate the reduction to the green channel.

Photoshop allows you to target noise reduction to a single channel.  You may be able to process the image twice in ACR, once with heavier noise reduction and once with less.  With both open in PS, replace the green channel from the less noise reduction with the one from the higher noise reduction.  

[kirkt]

Since neither plugin does much better on reducing the noise induced (or revealed?) by my shadows and exposure changes in ACR -- without killing detail --  I get the feeling that "my noise" isn't the same thing as the noise these plugins are designed for.  (I had assumed that the noise I'm fighting is still "device noise" from the D700, but just not evident in the original exposure.)

Here is a 300% screen shot of a portion of the image to show the noise I'm dealing with.  

What, specifically, is "your noise"?  That 300% crop seems pretty noise-free to me - that is, there are no chroma speckles or patterned noise or other artifacts that hide in shadows.  Did you produce this crop in ACR?  Was there any sharpening applied to this image (I assume there was no NR applied)?

What is your intended output and how does this noise manifest itself at that final output?  Proper exposure will give you smoother shadow tones, but the example you posted is not a big deal, typically.  Look at it as "grain" that takes the digital edge off of the image.

I have used Neat Image for years - like most NR applications, it will sample and construct a model of the noise present in an image and permit you to treat that noise according to its frequency and channel.  NR is a balancing act, suppressing just enough to eliminate artifacts while retaining detail.  "Your noise" would be sampled and accounted for in Neat Image - it is just a matter of dialing back the controls to a reasonable level.

kirk

[Guillermo Luijk]

I'm working in CS6 on some RAW images that were shot on a D700 at ISO200 and are a bit underexposed. When I open up shadows (+21) and increase exposure slightly (+1.65) in CS6 ACR noise is introduced.

I wanted to note that noise is not introduced by increasing exposure. Noise was there because of insufficient RAW exposure in that area of your image, you just made it visible by pushing the shadows.

Noise basically depends on the ISO set and the level of exposure achieved in the RAW file. The level of exposure achieved in the RAW file depends both on your overall exposure settings (aperture/shutter), and how dark the particular area of the scene was. You can ruin a portion of an image with noise because of insufficient overall exposure (we would be speaking of general underexposure here) and/or because of high dynamic range of the scene that made some part of it (the dark shadows) achieve very low levels in the RAW file.

Anyway I agree with Kirk, I don't see a specially noisy crop here. This degree of noise is typical and acceptable in the dark parts of any RAW file and shouldn't be an issue when printing or re-scaling the image for web.

Regards

[bjanes]

Profiles are built around normally/properly exposed images. In the case of D700 profile at 200 ISO, the profile would count on rather minimal, if any noise - that is why you do not see any effect on the image). The noise you are fighting is underexposed noise, which may (or not) have a different pattern than a "normal" noise (this is above my "pay grade" - we have experts here on the forum that might clarify this).

For practical purposes, do not rely on built-in profiles, but go manually.

Check this example from our member DigitalDog (the whole debate is in this thread):

http://digitaldog.net/files/100vs800iso.jpg

You will see that, under certain circumstances, an 800 ISO image might have less noise than a 100 ISO image, making a ready-made noise reduction profile practically useless.

Slobodan's comment about the noise profile for ISO 200 is critical. The ISO of the camera setting and the exposure correction are important components of the noise profile. Noise Ninja (see pp 46 and 49 of the user guide for details) incorporates the exposure correction in the raw converter as well as camera ISO in the profiles. For the ISO shot with +1.5 EV of exposure correction, one should use a profile made with these conditions.

It is best to minimize noise during capture and proper exposure is important as it reduces the effect of photon noise on the signal to noise ratio (the photon noise rises with exposure, but the SNR improves). Read noise is the other main source of noise (see Emil Martinec). With the D700, read noise at base ISO is high, and is reduced as one increases the ISO to 800 or thereabouts, so better results with this camera are obtained by increasing the ISO on the camera. With "ISO less" cameras such as the Nikon D7000 it makes little difference whether one increases the ISO on the camera or uses positive exposure compensation in the raw converter. One must know the characteristics of his/her camera.

Photon noise is present when the light hits the sensor and is not affected by the ISO setting of the camera, but is determined by the number of photons collected (camera exposure). With the D700, read noise is affected by the ISO setting. Whether or not the spectrum of read noise differs from photon noise would be of interest, but I have no information on this matter.

Regards,

Bill

[Guillermo Luijk]

Whether or not the spectrum of read noise differs from photon noise would be of interest, but I have no information on this matter.

Both follow gaussian-like distributions with no spatial correlation with adjacent pixels (this excludes pattern noise which is caused by the particularities of the readout).

Genuine RAW histograms:

Read noise (from a Canon dark frame): see how the 3 channels overlap, demonstrating there is no channel differentiation in terms of noise.


Photon noise (histogram represents read noise+photon noise but RAW exposure was chosen to make photon noise prevail): as usual the G channel gets the highest exposure, but R was close in this case because of the tungsten light used to light the subject (a white uniform wall).


Regards

[mouse]

 With the D700, read noise at base ISO is high, and is reduced as one increases the ISO to 800 or thereabouts, so better results with this camera are obtained by increasing the ISO on the camera.
Bill

Not sure how you reconcile this statement with this.
Click on D700_14

However I would certainly agree that increasing the iso (to 800 or more) results in a lot less noise than underexposing at base iso.

Best regards.

[bjanes]

Not sure how you reconcile this statement with this.
Click on D700_14

However I would certainly agree that increasing the iso (to 800 or more) results in a lot less noise than underexposing at base iso.

Best regards.

A good question and the answer is not that simple, but the concepts are important for exposing is light starved situations. Bill Claff's data are accurate, but he is measuring read noise in ADUs (raw data numbers) rather than in electrons. See Emil Martinec's section "S/N and Exposure Decisions" look at Figures 15a and 12a and read the accompanying text.

Roger Clark reaches a similar conclusion. See the Conclusion section of the linked article.

These seemingly paradoxical results are because with some cameras, the electronics are not well matched to the sensor. With improved electronics such as with the D7000, the ISO is no longer relevant with respect to the effects of read noise.

Regards,

Bill

[Dinarius]

Just found this topic via a search and, since it's an ongoing issue, I see no reason to start a new topic.

I am frequently in situations where I am faced with shooting long(ish) exposures (though never more than ISO 100/6 seconds) or raising the ISO on my 1Ds Mklll.

Typically, I shoot ISO 100, 1 to 2 seconds at f/11.

For reasons of sweet spot and/or depth-of-field, let's assume that the aperature cannot be changed....

Am I right in concluding based on what's been said above, that I am better off shooting, for example, 1/2 second, f/11, ISO 400, rather than 2 seconds, f/11, ISO 100, when it comes to minimising noise?

Thanks.

D.

[fdisilvestro]

Hi,

In your example you are keeping the correct exposure in both scenarios, (1/2 sec, f/11, ISO400 vs. 2 sec, f/11, ISO100) so no increased noise due to underexposure.

Considering the sensor performance information from DXOMARK for your camera model (Canon 1Ds MkIII), there is a minimal difference in image quality (~0.5 EV dynamic range loss) from shoting at ISO 100 or ISO 400, as long as you expose correctly (If you underexpose the ISO 100 capture, you will end up with more noise than using ISO 400)

So what to use? Considering that at longer exposure times there is an increased chance of thermal noise, I would go for the 1/2 sec, ISO 400 option.

[Tim Lookingbill]

Why not bump it up to ISO 800, 1/2sec ETTR so the results are brighter in post and correct there? Wonder if that would reduce noise level appearance.